Should I review wins too?

Sometimes, and less often than losses. You blunder in games you win — the one-in-13 error rate does not check the result — and those blunders are the ones you never notice because you got away with them. But losses give you more material per minute of attention. My split is roughly four losses to one win, which is a preference, not a finding.

Is fifteen minutes really enough?

Enough for what depends on you. Fifteen focused minutes a day on your own errors is more targeted study than most players under 1800 do in a week. Whether it moves your rating 100 points or 300 points in a year, nobody has measured well, and anyone who gives you a number for that is guessing.

What about tactics puzzles?

They are fine, and they are not the same thing. A puzzle tells you a tactic exists in the position. Your own games do not. Most of your blunders happen because you did not know it was a critical moment, and puzzles train the solving, not the noticing. Use puzzles as warm-up, not as the plan.

Do I need an engine at all, or can I just think about the game?

You need something to tell you the truth. If you had known the move was bad you would not have played it, so your own review will miss the same things twice. Any free engine analysis is enough — you are looking for evaluation cliffs, not for depth-40 accuracy.

← Blog

Build your chess improvement plan out of your own losses

A practical method for turning last week's losses into a fifteen-minute daily session, and what our engine data says about where your errors actually live.

3 September 2026·7 min read·Includes figures from Mated’s own move data

The plan: fifteen minutes, three of your own positions

Take the last five games you lost. Find the three moments where the evaluation swung hardest against you. Set each position up on a board, spend five minutes trying to find the move you missed, then check the engine. That is the whole plan. Repeat it daily with fresh games as they accumulate.

That is it because the alternative — a course, an opening repertoire, a tactics streak — is built for an average player who does not exist. Your errors have a shape. Someone else's course does not know that shape. Fifteen minutes on your own worst three positions is aimed at exactly the thing costing you rating points this month.

The rest of this article is about why three positions is enough, where in the game to look for them, and what to do when you find the same mistake three weeks running.

Five recent losses, three positions each session, five minutes per position.
Guess first, engine second. Reverse that order and you learn nothing.
Write down the reason you missed it in one sentence.
Move to new games when you stop being surprised by the old ones.

You make about one real error every 13 moves

Across 124,894 engine-classified moves in 4,019 analysed games on Mated, 3.8% of moves played are outright blunders and another 4.2% are mistakes. That is roughly one real error every 13 moves, before you count the 18.2% classified as inaccuracies.

That number should change how you plan. If you are erroring once every 13 moves, a 40-move game hands you three serious mistakes to study. Five games gives you fifteen. You do not have a shortage of material. You have a shortage of attention, and you are spending it on other people's material.

It also means the honest ceiling on your improvement is not knowledge. You are not losing to a gap in your Sicilian theory when one move in 13 is a mistake or worse. You are losing to whatever is producing those moves — and that cause is visible in your games and invisible in any course.

The middlegame is where the damage is, so that is where to look

In the same 124,894-move sample, 9.0% of middlegame moves are a mistake or worse. In the endgame it is 2.3%. Errors cluster where the position is most complicated and least memorised.

So when you go looking for your three positions, start after the opening and stop before the queens come off. That is a bias, not a rule — if you keep losing rook endings you should study rook endings — but it is the right default. It also means the most common self-directed chess improvement plan, which is "learn more openings", is aiming at the wrong part of the game.

One caveat on the endgame figure: endgames have fewer plausible moves, and an engine calls a move fine as long as it does not throw the win. A 2.3% error rate does not mean you play endgames well. It means you rarely play them catastrophically. Those are different things and this data cannot separate them.

A small number of terrible moves carries most of the loss

The average move in that sample gives away 95 centipawns against the engine's best. The median is far lower. The gap between those two numbers is the whole story: a handful of very bad moves is doing most of the damage, and the ordinary moves in between are broadly fine.

This is good news for a fifteen-minute plan. You are not trying to raise the quality of every move you make, which would be hopeless. You are trying to stop three or four disasters per game. That is a narrow target and it is reachable.

It also tells you what to ignore. When the engine tells you a move was 0.3 worse than the best move, close the tab. That is not the move that lost you the game. Look for the swing of a piece or more. If you cannot find one, you probably lost slowly to a plan you did not understand, which is a different kind of study session.

Sort by evaluation swing, not by move order.
Ignore anything under about half a pawn.
One 4.0 swing matters more than twenty 0.2 swings.

The one-sentence diagnosis

Finding the blunder is easy. Naming why you played it is the part people skip, and it is the part that transfers to the next game.

Here is a worked example. You are Black in a Queen's Gambit Declined structure, your bishop sits on d6, and you play a natural-looking ...c5 to free the position. The engine says it loses a pawn to a discovered attack you did not see because your eyes were on the queenside. The blunder is ...c5. The diagnosis is not "I missed a discovery." It is "I played a freeing move without checking what happened to my king's diagonal first."

Write that sentence down. Do it for three positions a day for two weeks and you will have thirty sentences. Read them. Most people find four or five phrasings repeating. Those repeats are your actual improvement plan — not the thirty positions, the five sentences.

I have no data on how many players do this and improve faster. It is the part of my own routine I would give up last, and I am telling you it is opinion.

What this costs you to set up, and where a tool helps

You can do all of this free. Download your games from Chess.com or Lichess, run the built-in analysis, scan the evaluation graph for the cliffs, and set those positions up on a physical board. That is maybe ten minutes of admin a session on top of the fifteen minutes of thinking. Many people are fine with that and should not pay anyone anything.

The admin is the part that kills the habit, though. Not the thinking — the finding. Mated exists for that: it pulls your recent games through the public Chess.com and Lichess APIs, runs Stockfish over every position in your browser, and hands you a fifteen-minute session assembled from your own worst moments rather than a puzzle set. If your problem is that you have never once opened your analysis tab, a tool that does the sorting is worth more to you than better advice.

If your problem is that you look at the analysis and shrug, no tool fixes that. That is the one-sentence diagnosis, and you have to write it yourself.

When to break the plan

Stop reviewing your own games when they stop producing new sentences. If your notes have said "I moved before checking my opponent's last move" fourteen times, you do not need a fifteenth position. You need to change something about how you play — a physical habit, a slower time control, a rule about not moving until you have said your opponent's threat out loud.

That is the limit of any plan built from your own losses. It is excellent at telling you what is broken and silent on how to fix a habit. At that point go find the specific resource: a book on rook endings, a coach, a set of games in your structure. Then come back to your losses to check whether it worked.

The loop is diagnose, treat, re-measure. Most chess advice sells you the treatment without ever doing the diagnosis, and then you cannot tell whether it worked.

The useful output of this routine is not the positions you solve. It is the short list of sentences describing how you lose, and that list is only about five items long — which is why a plan written for nobody in particular cannot contain it.

Questions

Should I review wins too?: Sometimes, and less often than losses. You blunder in games you win — the one-in-13 error rate does not check the result — and those blunders are the ones you never notice because you got away with them. But losses give you more material per minute of attention. My split is roughly four losses to one win, which is a preference, not a finding.
Is fifteen minutes really enough?: Enough for what depends on you. Fifteen focused minutes a day on your own errors is more targeted study than most players under 1800 do in a week. Whether it moves your rating 100 points or 300 points in a year, nobody has measured well, and anyone who gives you a number for that is guessing.
What about tactics puzzles?: They are fine, and they are not the same thing. A puzzle tells you a tactic exists in the position. Your own games do not. Most of your blunders happen because you did not know it was a critical moment, and puzzles train the solving, not the noticing. Use puzzles as warm-up, not as the plan.
Do I need an engine at all, or can I just think about the game?: You need something to tell you the truth. If you had known the move was bad you would not have played it, so your own review will miss the same things twice. Any free engine analysis is enough — you are looking for evaluation cliffs, not for depth-40 accuracy.

See this in your own games

Mated reads your last games, scores them across five categories and builds a fifteen-minute session out of what it finds. No card.

Read my games